What does Error Bars in Dating conclude, in numbers, about “The Expected Loss of Testing: When to Test and What a Test Can Answer”?

Answer: Chapter 6 writes testing as a problem of expected loss; under illustrative parameters a buyer should not buy untested below a prior of 0.655, and a test is worth most at middling priors.

Scope of the numbers: Objects, prices and parameters are illustrative, the calibration curve is a hand-set illustrative line rather than IntCal data, and all calculations are generated by the book's programs. Chapter 6 writes testing as a problem of expected loss; under illustrative parameters a buyer should not buy untested below a prior of 0.655, and a test is worth most at middling priors.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. standard-library Python onlyprinted values that match the answer: 0.655

price 100.0, worth 150.0 / 5.0, test cost 4.0, se 0.95, sp 0.9
prior pno testwith testtest valueloss vs perfect
0.100.00-4.00-4.005.00
0.300.003.603.6011.40
0.500.0015.0015.0010.00
0.706.5026.4019.908.60
0.9035.5037.802.307.20
0.9745.6541.79-3.862.85
break-even prior without test: 0.655
Show the full program
# Chapter 6. p61: should a piece be sent for testing before purchase?
# Expected-value decision with a prior p that the piece is of the age
# claimed. Asking price PRICE; worth V if the age claim holds, R if
# not. A test costs C, passes a true piece with probability SE and
# fails a wrong one with probability SP (a TL test cannot see every
# kind of wrong piece, so SP < 1). Without the test the buyer buys
# iff p V + (1 - p) R > PRICE. With it, he buys after a pass iff the
# posterior makes buying worth it. Value of the test = best with
# test - best without. All money figures are illustrative units.
import math

PRICE, V, R, C = 100.0, 150.0, 5.0, 4.0
SE, SP = 0.95, 0.90


def no_test(p):
    return max(0.0, p * V + (1.0 - p) * R - PRICE)


def with_test(p):
    branches = []
    for good, bad in ((p * SE, (1.0 - p) * (1.0 - SP)),
                      (p * (1.0 - SE), (1.0 - p) * SP)):
        branches.append(max(0.0, good * (V - PRICE)
                            + bad * (R - PRICE)))
    return math.fsum(branches) - C


def perfect(p):
    return p * (V - PRICE)


if __name__ == "__main__":
    print("price {}, worth {} / {}, test cost {}, se {}, sp {}".format(
        PRICE, V, R, C, SE, SP))
    print("  prior p  no test  with test  test value  loss vs perfect")
    for p in (0.10, 0.30, 0.50, 0.70, 0.90, 0.97):
        a, b = no_test(p), with_test(p)
        best = max(a, b)
        print("  {:<7.2f}  {:>7.2f}  {:>9.2f}  {:>10.2f}  {:>15.2f}"
              .format(p, a, b, b - a, perfect(p) - best))
    pe = (PRICE - R) / (V - R)
    print("break-even prior without test: {:.3f}".format(pe))

Single source and citation

Fan, Y. (2026). Error Bars in Thermoluminescence and Radiometric Dating: Multiple Solutions on Calibration Plateaus and When to Send Antiques for Testing (Version 1.0.0), Chapter 6: The Expected Loss of Testing: When to Test and What a Test Can Answer. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-331 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026331,
  author    = {Fan, Yafu},
  title     = {Error Bars in Thermoluminescence and Radiometric Dating: Multiple Solutions on Calibration Plateaus and When to Send Antiques for Testing},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book